A first look at Solaris performance: prstat, vmstat, mpstat, iostat and zpool iostat

What it is

Five read-only observability commands cover most first-look performance questions on Solaris 11.4: prstat for per-process/per-user CPU and memory, vmstat for kernel, virtual-memory and CPU summary statistics, mpstat for per-processor statistics, iostat for disk I/O, and zpool iostat for per-pool and per-device ZFS I/O. None of them change system state; all are safe to run repeatedly on a production host.

Why it matters

Solaris's process and I/O model differs enough from Linux's that the similarly-named Linux tools are not simply "the same tool, different packaging" — flag meanings and output columns differ, and Solaris adds zone-awareness that has no equivalent in the same Linux tools.

How to apply

• Get a live, top-like process view: prstat alone refreshes periodically; -a additionally displays separate reports about processes and users at the same time; -Z displays separate reports about processes and zones at the same time, which matters on a host running multiple non-global zones where one zone's noisy workload can otherwise be invisible in the aggregate.
• Check overall system pressure: vmstat <interval> repeated at an interval reports kernel thread, virtual memory, disk, trap and CPU activity; the first line is a since-boot summary, so always look at the second and later lines for current activity.
• Break CPU usage down per processor: mpstat <interval> reports processor statistics in tabular form, one row per processor (or processor set), useful for spotting one hot CPU that an aggregate vmstat view would average away.
• Check disk-level latency and throughput: iostat -xnz <interval> — -x reports extended disk statistics, including wsvc_t (average time in the wait queue), asvc_t (average service time of active transactions) and %b (percent busy); -n displays device names in descriptive format instead of raw instance names; -z omits lines whose underlying data values are all zero, keeping the output focused.
• Check ZFS pool I/O specifically, since plain iostat reports on the underlying devices rather than on pool-level behaviour: zpool iostat -v <pool> <interval> shows read/write operations and bandwidth per top-level device.

Pitfalls

• Reading the first report of vmstat, iostat or mpstat as current activity: it is a since-boot summary, not a snapshot; only the second and later reports cover the preceding interval.
• Running prstat without -Z on a host with non-global zones and concluding a zone is idle because its processes do not stand out among the global list; pass -Z explicitly to check.
• Comparing asvc_t directly against application-level latency: it covers only time being serviced by the device; wsvc_t adds the driver's wait queue, and neither includes time spent in the file system or application above the driver.